Soils and Foundations 2015;55(2):437–446

HOSTED BY The Japanese Geotechnical Society

Soils and Foundations

www.sciencedirect.com
journal homepage: www.elsevier.com/locate/sandf

Reliability-based analysis of combined installation damage and creep for the

tensile rupture limit state of geogrid reinforcement in Japan
Richard J. Bathursta,1, Yoshihisa Miyatab,n
a
Geo-Engineering Centre at Queen's-RMC, Department of Civil Engineering, Royal Military College of Canada, Kingston, Ontario, Canada K7K 7B4
b
Department of Civil and Environmental Engineering, National Defense Academy, 1-10-20 Hashirimizu, Yokosuka 239-8686, Japan
Received 12 May 2014; received in revised form 29 September 2014; accepted 21 December 2014
Available online 16 March 2015

Abstract

The paper uses statistical data for the prediction of installation damage and creep-reduced strength collected by the writers in earlier
investigations to estimate the probability of failure of tensile rupture of geogrid reinforcement products. The original data were compiled from
Public Works Research Center (PWRC) geogrid product certification reports issued in Japan. The paper develops the formulation for the ultimate
tensile rupture limit state equation and links it to allowable stress design (ASD) practice currently used in Japan and reliability theory-based load
and resistance factor design (LRFD) used in North America. The paper shows that variability in the prediction of creep-reduced strength is largely
captured by the inherent variability in strength of the materials at the time of manufacture. Combined variability due to creep and installation
damage is typically dominated by variability in the prediction of strength after installation damage. Where this is not the case the combined
variability is very low (less than 5%). The variability in the estimate of strength reduction due to combined installation damage and creep is
demonstrated to be less than the variability in the estimates of reinforcement load even for the case of a load model judged to give relatively
accurate load predictions. For poorer load models the under-prediction of reinforcement loads provides an additional margin of safety. The paper
provides a framework for future rigorous reliability theory-based LRFD calibration for the ultimate tensile rupture of geogrid reinforcement in
reinforced soil applications in Japan and elsewhere, and provides the necessary bias statistics for the resistance side in the ultimate tensile rupture
limit state equation.
& 2015 The Japanese Geotechnical Society. Production and hosting by Elsevier B.V. All rights reserved.

Keywords: Geogrid; Installation damage; Creep; Reliability; Load and resistance factor design; LRFD; Factor of safety; Tensile rupture; Limit state

1. Introduction

Abbreviations: AASHTO, American Association of State Highway and
Transportation Officials (USA); ASD, allowable stress design; ASTM,
American Society for Testing and Materials (USA); COV, coefficient of
variation ( ¼standard deviation/mean); HDPE, high-density polyethylene;
PWRC, Public Works Research Center (Japan); LRFD, load and resistance
factor design; PET, polyester; POM, polyoxymethylene; PP, polypropylene
n
Corresponding author. Tel.: þ81 468 41 3810; fax: þ 81 468 44 5913.
E-mail addresses: bathurst-r@rmc.ca (R.J. Bathurst),
miyamiya@nda.ac.jp (Y. Miyata).
1
Tel.: þ1 613 541 6000x6479/6347/6391; fax: þ1 613 541 6218.
Peer review under responsibility of The Japanese Geotechnical Society.
An allowable stress design (ASD) approach is currently used
in Japan to calculate the long-term design strength of geosyn-
thetic reinforcement layers in reinforced soil walls, slopes and
embankments (Public Works Research Center
PWRC,
2013). The design tensile load is calculated as Tdes ¼ FTmax
where the maximum tensile load in a layer (Tmax) is multiplied
by a minimum specified factor of safety (F) for each limit state
(e.g. F ¼ 1 and 1.5 for tensile rupture in walls and embank-
ments, respectively, and F ¼ 2 for pullout). The design tensile

http://dx.doi.org/10.1016/j.sandf.2015.02.017
0038-0806/& 2015 The Japanese Geotechnical Society. Production and hosting by Elsevier B.V. All rights reserved.
438 R.J. Bathurst, Y. Miyata / Soils and Foundations 55 (2015) 437–446
load is assumed to act for the life of the structure and cannot
exceed the long-term allowable strength of the reinforcement
(Tal). Hence, in ASD practice the ultimate tensile rupture limit
state of a reinforcement layer is
T al Z FT max ð1Þ
An alternative method for design against rupture in rein-
forced soil wall structures is the load and resistance factor
design (LRFD) framework used in North America (AASHTO,
2012; CSA, 2006). The equivalent design equation for the
ultimate (tensile) rupture limit state of a reinforcement layer
subjected to a single load term (e.g. load due to soil self-weight
in a reinforced soil wall application) can be expressed as
φT al Z γ Q T max ð2Þ
Here φ is the resistance factor and γQ is the load factor. The
expectation is that design outcomes will have a probability of
failure that is acceptable (i.e. small) when the inequality is
satisfied. An important constraint on the factors in Eq. (2) is
that φ r 1 and γQ Z 1 to be consistent with LRFD practice.
Regardless of the design approach (ASD or LRFD), the
computation of the (nominal) long-term allowable strength
available at the end of design life is computed as
T ult T ult
T al ¼ ¼ ð3Þ
RF RFCR  RFID  RFD  RFJ
Here, Tult is the in-isolation ultimate tensile (reference)
strength of the reinforcing geosynthetic expressed in units of
force per unit width of material. RF is the product of reduction
factors to account for potential strength loss over the design
life of the structure due to installation damage (RFID), creep
(RFCR), chemical/biological degradation processes (i.e. dur-
ability) (RFD), and connections (junctions) (RFJ).
The focus of this paper is the influence of variability in the
prediction of installation damage and creep-reduced strength of
geogrid reinforcement products on the probability of failure for
the ultimate tensile rupture limit state. The limit state is
expressed in a load and resistance factor design (LRFD)
format. The data for the current study have been taken from
two complementary investigations by the writers (Miyata et al.,
2014; Miyata and Bathurst, 2015). They reviewed PWRC
product certification reports to calculate statistical variations in
the predictions of strength loss due to installation damage and
creep for geogrid products used in Japan. The general
approach used in this paper to estimate probability of failure
of the ultimate tensile rupture limit state follows that described
by Bathurst et al. (2011c). However, in this prior work, creep
bias statistics were not available and bias values for installation
damage were computed using data from western installation
damage testing protocols which are different from the Japanese
methodology (Miyata and Bathurst, 2015). Another unique
feature of the current investigation is a quantitative assessment
of the influence of under-estimation of the reference tensile
strength (Tult) of a geogrid reinforcement material on prob-
ability of failure of the tensile rupture limit state.
2. LRFD tensile rupture limit state equation incorporating
variability in strength reduction contributions
The limit state equation (performance function) for the long-
term tensile rupture of a reinforcement layer is expressed as
g ¼ T al;meas –T max ;meas ð4Þ
where Tal,meas is a random variable representing the measured
long-term tensile strength of a reinforcement layer and Tmax,
meas is a random variable representing the maximum measured
tensile load in the same layer. As examples, the layer could be
a reinforcement layer in a geosynthetic reinforced soil wall or
embankment. The probability that this limit state is less than
zero, denoting failure due to long-term rupture (i.e. random
variable g o 0), can be equated to variability in measured load
and measured resistance (strength) values. In the developments
to follow this variability is quantified by the mean and spread
of bias values, where bias is defined as the ratio of measured
value to predicted (nominal) value (Allen et al., 2005; Bathurst
et al., 2008, 2011a, 2011b, 2011c; Bathurst, 2014). In this
paper the following nomenclature is used for the load bias:
X Q ¼ T max ;meas =T max ð5Þ
In the context of geotechnical soil-structure design, the
magnitude of load bias values will depend on model accuracy
(the intrinsic accuracy of the deterministic theoretical, semi-
empirical or empirical model representing the mechanics of the
limit state under investigation), random variation in input
parameter values, spatial variation in input values, quality of
data and, consistency in interpretation of data when data are
gathered from multiple sources, which is the typical case
(Allen et al., 2005).
The resistance bias is expressed as
X R ¼ T al;meas =T al ð6Þ
Here Tal,meas is measured tensile rupture strength and Tal is the
predicted value using Eq. (3).
In the context of reinforced soil structures, resistance bias is
a measure of the variability of actual available strength with
respect to the nominal value used in the limit state design
equation (i.e. Eqs. (1) and (2)).
Substituting bias terms into Eq. (4) gives
g ¼ T al X R –T max X Q ð7Þ
An important condition to allow this substitution to be made
is that bias ratios and predicted (nominal) values in the
denominator are independent (uncorrelated) (e.g. Bathurst
et al., 2008, 2011a; Bathurst, 2014). Strategies to remove
hidden dependences (correlations) include assigning different
load or resistance factors to different ranges of bias values or to
modify the underlying load or resistance model to ensure that
bias values do not vary with magnitude of the nominal value to
an acceptable significance level (Bathurst et al., 2008, 2012b;
Huang and Bathurst, 2009).
 R.J. Bathurst, Y. Miyata / Soils and Foundations 55 (2015) 437–446
From Eq. (3) it follows that (Bathurst et al., 2011c)
T ult
T al X R ¼ XR
RF


 RF
RF D  RF
J 

ID CR
T ult 1 1 1
¼ X ID X CR XD XJ
RFID RFCR RFD RFJ
The bias of the long-term tensile strength of the reinforce-
ment, XR, is expressed as a product of four bias values:
X R ¼ X ID X CR X D X J
where XID is bias of tensile strength after installation damage,
XCR is bias of tensile strength after creep, and XD and XJ are
bias of strength values after degradation processes and due to
any connections (junctions), respectively. Assuming that XID,
XCR, XD and XJ are uncorrelated, the mean and coefficient of
variation (COV) of XR values are given by Eqs. (10) and (11),
respectively (Ang and Tang, 1975):
μR ¼ μX ID μX CR μX D μX J
and
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
COV R ¼ COV 2X ID þ COV 2X CR þ COV 2X D þ COV 2X J
Here, the subscripts for each of the mean and COV terms can
be matched to installation damage, creep, degradation and
junction reduction factors. In theory, Eqs. (10) and (11) are
valid only for uncorrelated log-normal distributions of the four
random variables. However, for uncorrelated normal distribu-
tions with small coefficients of variation these expressions are
sufficiently accurate for reliability analyses (Ang and Tang,
1975). In the current study, the contribution of variability in
the prediction of reduced strength due to durability and
junctions to XR is ignored since these data are not available.
Therefore, μX D ¼ μX J ¼ 1 and COVX D ¼ COVX J ¼ 0:
The minimum value of Tal to satisfy Eq. (2) is
γQ
T al ¼ T max ð12Þ
φ strength after reduction for installation damage. The quantity
This condition corresponds to g ¼ 0 in Eq. (7) and delineates
between safe and unsafe outcomes. In fact, the factor of safety
used in ASD practice can be understood to be F ¼ γQ/φ.
Substituting into Eq. (7) gives
γQ
g¼ XR
XQ ð13Þ
φ T ult;meas
However, recall that Tal includes four reduction factors
(RF ¼ RFCR  RFID  RFD  RFJ ) which results in the (nom-
inal) long-term allowable strength (Tal) at the end of design life
being (typically) much lower than the reference tensile strength
(Tult) of the reinforcement. Hence, the use of the terms
“failure”, “unsafe”, and “probability of failure” does not imply
structure collapse; rather, failure means Tal in Eq. (3) is less
than the factored load (FTmax) applied to the reinforcement
layer.
LRFD calibration for the limit state for long-term tensile
rupture of a geosynthetic reinforcement layer can be under-
stood to be the selection of load and resistance factors so that
the probability that g is less than zero [Pf(g o 0)] in Eq. (13)
does not exceed a prescribed value. The probability of failure
439
is related to the reliability index β as follows:
Pf ¼ 1
ΦðβÞ ð14Þ
Here, Φ is the standard normal cumulative distribution func-
ð8Þ tion (NORM.DIST in Excel). A useful closed-form solution for
the calculation of the reliability index is

r ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  

ln ðγ Q =φÞðμR =μQ Þ 1 þ COV 2Q = 1 þ COV 2R
β¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
h  ð15Þ
ð9Þ i
ln 1þ COV 2Q 1 þ COV 2R
For any trial combination of load and resistance factor, and
knowing the mean (μQ and μR) and coefficient of variation
(COVQ and COVR) of the load and resistance bias values, the
probability of failure can be computed. An alternative and
more robust strategy to compute the probability of failure is to
use Monte Carlo simulation. The advantage of Monte Carlo
ð10Þ simulation is that assumed distributions for bias values are
unrestricted. The advantage of Eq. (15) is that it can be used to
quickly generate smooth distributions of β (or Pf) versus γQ/φ
ð11Þ which are useful for sensitivity analysis and presentation of
results (Bathurst et al., 2011c). In the next section the
methodology to compute the bias statistics for installation
damage and creep-reduced strength is outlined.
3. Calculation of bias statistics
The variability of reinforcement strength immediately after
installation can be quantified by the bias value XID introduced
previously and computed as (Bathurst et al., 2011c; Miyata
and Bathurst, 2015)
X ID ¼ T ID;meas =T ID ð16Þ
Here, TID,meas is measured tensile strength of a damaged
specimen and TID=T ult;meas /RFID is predicted (nominal)
T ult;meas is the average of tensile strengths (Tult,meas) from
multiple undamaged specimens. For example, the PWRC
(2013) guidance document requires a minimum of five tests
to compute the mean strength of undamaged specimens. RFID
is the installation damage reduction factor computed as
RFID ¼ ð17Þ
T ID;meas
Here, the denominator (T ID;meas ) is the mean value of the
tensile strengths of damaged test specimens (a minimum of
five tests).
Variability in undamaged tensile strength is expressed in the
same manner using the undamaged bias value computed as
X ult ¼ T ult;meas =T ult;meas ð18Þ
An example of the calculation of bias statistics for tensile
strengths for undamaged and damaged specimens is given by
Miyata and Bathurst (2015).
Miyata et al. (2014) quantified variability in the prediction
of creep-reduced strength using the creep bias value XCR.
Creep bias values were computed as the ratio of measured
440 R.J. Bathurst, Y. Miyata / Soils and Foundations 55 (2015) 437–446

Table 1
Summary of PWRC database for installation and creep testing.

Data Geosynthetic Number Manufacturer Range of undamaged Computed range of RFID (all Creep reduction factor PWRC Certification
group type of identification tensile strength Tulta values; Miyata and Bathurst, RFCRb (Miyata et al., report numberc
products (kN/m) 2015) 2014) year

1 Uniaxial 1 Mitsui Sanshi 38 1.19–1.74 2.19 2006 0901

HDPE Co.
2 Uniaxial 3 Mitsui Sanshi 43–89 1.12–1.43 1.98 2006 0901
HDPE Co.
3 Uniaxial 4 Mitsui Sanshi 115–192 1.08–1.26 1.80 2006 0901
HDPE Co.
4 Uniaxial 3 Geosystem Co. 45–108 1.05–1.17 2.10 2004b 0413
HDPE
5 Uniaxial 1 Bridgestone Co. 69 1.28–1.61 2.73 2000b 1110
HDPE
6 Uniaxial 2 Bridgestone Co. 91–107 1.00–1.09 2.73 2000b 1110
HDPE
7 Uniaxial PP 1 Takiron Co. and 268 1.00–1.11 1.46 2007a 0214
taped Asahi-Kasei Co.
8 Woven and 2 Toyobo Co. and 44–84 1.04–1.11 1.62 2008b 0808
knitted PETe Daika Co.
9 Woven and 1 Toyobo Co. and 159 1.04–1.10 1.62 2008b 0808
knitted PETe Daika Co.
10 Aramidf 1 Maeda Co. and 150 1.04–1.10 1.67 2002b 0120
Zeon Co.
f
11 Aramid 3 Maeda Co. and 37–196 1.03–1.05 1.51 2008a 0804
Zeon Co.
12 POMg 4 Kuraray Co. and 57–113 1.03–1.15 2.18 2004a 0412
Taiyokogyo Co.
Σ¼ 26

Using data from PWRC reports. T ult ¼ T ult;meas ¼ average rupture strength of five test specimens. Test specimen width¼200 mm and tests run at 1% strain/
a

minute.
b
Applied to reference tensile tests run at 1% strain/minute. Assuming design life of 120 years.
c
Certification number is used here to identify the matching PWRC report in the reference list.
d
Heat-bonded junctions and coated with Vinylon.
e
Polypropylene (PP) coating.
f
Aramid filament bundles coated with polyethylene.
g
Polyoxymethylene fibers coated with ethylene–vinyl acetate polymer (Akagi et al., 2004).

creep-reduced strength to nominal value. The nominal
value was computed from an equation fitted to the master
stress
rupture curve for geogrid products taken from the same
product line using the method by Bathurst et al. (2012a).
Finally, it should be noted that the formulation for calcula-
tion of bias values results in the mean bias value equal to one.
4. Database of installation and creep tests
4.1. General
Table 1 gives a summary of available test data from PWRC
product certification reports. Only data available for both
installation damage and creep testing on the same products
were used in the current study. The writers were given access
to PWRC archives to find any missing data. The same series
of PWRC product certification reports have been used by
the writers for statistical analysis of the accuracy of current
and proposed models for geogrid-soil pullout (Miyata and
Bathurst, 2012), geogrid creep (Miyata et al., 2014) and
installation damage (Miyata and Bathurst, 2015).
4.2. Undamaged and damaged bias statistics
A summary of bias statistics for tensile strength before and
after installation damage for different data groups is presented
in Table 2. A detailed discussion of the results in this table can
be found in the companion paper by Miyata and Bathurst
(2015) and for brevity is not repeated here. They concluded
that differences in COV values less than 1% are at the limit of
practical measurement. When all test data from the installation
damage testing were considered as a single population, there
was a positive correlation between COVX ID and RFID. How-
ever, correlations varied widely between data groups. Max-
imum computed COV values for installation damage bias (XID)
are shown in Table 3. These values correspond to the largest
computed value of COVX ID from any of the products in the
data group.
Fig. 1a shows the values of COVX ID computed for each
product in the installation damage database plotted against the
corresponding COVX ult using the database compiled by Miyata
and Bathurst (2015). The data show that in most cases the
variability in tensile strength after installation damage for
 R.J. Bathurst, Y. Miyata / Soils and Foundations 55 (2015) 437–446 441

Table 2
Bias values from tensile testing of undamaged and damaged samples (Miyata and Bathurst, 2015).

Data Geosynthetic Undamaged samples Damaged samples

group type
Number of Xult range COVX ult Number of Computed range of RFID XID range COVX ID
specimens specimens (all values)
Soil Type 1 Soil Type 2 Soil Type 3
D50 ¼21 mm D50 ¼ 11 mm D50 ¼ 1 mm

1 Uniaxial HDPE 5 0.99–1.01 0.006 15 1.19–1.74 0.87–1.10 0.095 0.049 0.062

2 Uniaxial HDPE 15 0.97–1.02 0.012 45 1.12–1.43 0.84–1.18 0.094 0.040 0.069
3 Uniaxial HDPE 20 0.98–1.01 0.008 60 1.08–1.26 0.92–1.09 0.043 0.042 0.023
4 Uniaxial HDPE 15 0.98–1.02 0.010 45 1.05–1.17 0.84–1.14 0.080 0.067 0.024
5 Uniaxial HDPE 5 0.99–1.01 0.007 15 1.28–1.61 0.88–1.17 0.111 0.051 0.015
6 Uniaxial HDPE 10 0.99–1.01 0.007 30 1.00–1.09 0.90–1.07 0.065 0.012 0.017
7 Uniaxial PP tape 5 0.97–1.02 0.023 15 1.00–1.11 0.96–1.07 0.033 0.051 0.023
8 Woven and 10 0.98–1.02 0.012 30 1.04–1.11 0.99–1.01 0.006 0.006 0.010
knitted PET
9 Woven and 5 0.98–1.02 0.015 15 1.04–1.10 0.99–1.01 0.007 0.007 0.007
knitted PET
10 Aramid 5 0.97–1.04 0.027 15 1.04–1.10 0.94–1.04 0.017 0.026 0.019
11 Aramid 15 0.94–1.05 0.026 45 1.03–1.05 0.92–1.06 0.017 0.034 0.018
12 POM 20 0.98–1.02 0.009 60 1.03–1.15 0.90–1.10 0.063 0.032 0.026
Σ ¼ 130 390

Notes: Mean of undamaged and damaged bias values μX ult ¼ μX ID ¼ 1.

Table 3
Combined COV of bias values XR ¼XIDXCR.

Data Geosynthetic type Range of tensile strength (kN/m)b Coefficient of variation of bias values
group
Adjusted from 1% From T ult;meas Tult recommended by Installation Creepc Combinedd
strain/mina database manufacturer damage
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
10% strain/min 1% strain/min Maximum Maximum COVR ¼ COV2X ID þ COV2X CR
COVX ID COVX CR

1 Uniaxial HDPE 45 38 36 0.095 0.006 0.095

2 Uniaxial HDPE 51–105 43–89 40–80 0.137 0.006 0.138
3 Uniaxial HDPE 135–224 115–192 100–160 0.063 0.006 0.063
4 Uniaxial HDPE 52–132 45–108 36–94 0.124 0.006 0.124
5 Uniaxial HDPE 77 69 55 0.111 0.006 0.111
6 Uniaxial HDPE 115–135 91–107 85–101 0.086 0.006 0.086
7 Uniaxial PP tape 273 268 220 0.011 0.034 0.036
8 Woven and 46–88 44–84 38–74 0.011 0.048 0.049
knitted PET
9 Woven and 165 159 144 0.007 0.048 0.049
knitted PET
10 Aramid 151 150 140 0.026 0.035 0.043
11 Aramid 39–202 37–196 34–185 0.052 0.023 0.057
12 POM 58–114 57–113 53–102 0.095 0.070 0.118
a
See Miyata and Bathurst (2007).
b
Test specimen width¼ 200 mm.
c
Data from Miyata et al. (2014).
d
Mean of undamaged and damaged bias values μX ult ¼ μX ID ¼1.

HDPE, PP and POM products is greater than for the
undamaged samples. Hence, installation damage adds to the
spread in strength values beyond the variability in virgin
strength values that occurs at the time of manufacture. For the
PET and Aramid materials there is no visual difference,
suggesting that the spread in strength values for these materials
is less affected by damage, at least under the test conditions
prescribed in the field installation damage test protocol
(PWRC, 2013).
4.3. Creep bias statistics
Maximum computed COV values for creep bias (XCR) are
shown in Table 3 using the compiled database reported in the
442 R.J. Bathurst, Y. Miyata / Soils and Foundations 55 (2015) 437–446
Fig. 1. Variability in installation damage bias and creep bias versus strength bias of undamaged (virgin) test samples. Note: Geosynthetic product types are the same
in both figures but data are from different products or samples and the number of data points for each product is different in each database. (a) Installation damage
(data from Miyata and Bathurst, 2015). (b) Creep (data from Miyata et al., 2014).
companion paper by Miyata et al. (2014). The data are for the
same or similar products in the installation damage database
(Miyata and Bathurst, 2015). Bias values were computed as the
ratio of measured strength to predicted strength for each data
point. The predicted (nominal) strength was computed using
the equation for the creep-rupture curve fitted to all data points
in each data group for times to rupture up to 1000
10,000 h.
Individual COVX CR values for different products are plotted
against COVX ult in Fig. 1b. Miyata et al. (2014) concluded that
for those data with multiple data points there is no practical
systematic difference in spread of bias values for creep-
reduced strength and virgin strength of these materials. The
exception is the HDPE data; however, the COV values for
creep and virgin strength bias in this case are very low and
difficult to detect. Taken together the spreads in creep-reduced
strength are low and can be assumed to be due to inherent
variability in the strength of the virgin materials at time of
manufacture. As a point of reference, consider that the COV of
the rupture strength of SS400 reinforcement steel in Japan is
5% (Tanaka and Sakai, 1979). The creep-reduced strength bias
values in Fig. 1b are 7% or less (and most often less than 5%)
which is very small when compared to the prediction accuracy
of load-side contributions to the tensile rupture limit state
function discussed in Section 5.
4.4. Influence of tensile test strain rate and other factors
Also presented in Table 3 are reference tensile strength
ranges for the data groups used in the current study.
In Japanese practice, undamaged and installation damaged
specimens are tested in air at 20% strain/minute. The calcula-
tions of RFID and bias statistics are based on a ratio of strength
values so the rate of testing does not influence damage bias
statistics. However, the reference strength Tult used in design
(Eq. (1)) is based on the same tensile test method but from tests
run at 1% strain/minute (PWRC, 2013). Many geosynthetic
reinforcement materials have mechanical properties that are
strain-rate dependent, particularly polyolefin-based (HDPE and
PP) geogrids (e.g. Bathurst and Cai, 1994; Walters et al., 2002;
Hirakawa et al., 2003; Kongkitkul et al., 2004, 2007a; Shinoda
et al., 2002; Shinoda and Bathurst, 2004a, 2004b). Miyata and
Bathurst (2007) calculated a set of factors for the product types
used in the current study that can be used to convert rupture
loads (Tult) between tests carried out at strain rates of 1%, 10%
and 20% strain/minute. Table 3 also shows rupture strengths
for the data groups in this study assuming that reference tensile
strengths are from tests run at 10% strain/minute, which is the
rate used in North American practice (ASTM D6637, 2001).
Hence, this table can be used to select COV values for
installation damage bias for probabilistic analysis of geogrid
reinforced structures where reference design strengths are
based on the faster ASTM tensile test protocol.
Another practical consideration for both ASD and LRFD is
that manufacturers do not always use reference strengths
computed from an average of tests as was done in the previous
related analyses by the writers (Miyata et al., 2014; Miyata and
Bathurst, 2015). Typically, a manufacturer will report a lower
ultimate tensile strength value in order to a) position the
product in the market, b) consider variation in reported results
between laboratories, or c) to be cautious (Bathurst et al.,
2012a). The manufacturer recommended Tult values for design
that appear in the PWRC certification reports are shown in
Table 3. The ratio of the true mean value (T ult;meas ) computed
for the same product (or product line) to the recommended Tult
value for design (Eq. (1)) ranges from 1.05 to 1.35 but varies
widely between product types and even within product lines.
 R.J. Bathurst, Y. Miyata / Soils and Foundations 55 (2015) 437–446
The practice of using a conservative estimate of Tult for design
will reduce the probability of tensile rupture failure as
illustrated quantitatively in the next section.
5. Calculation of probability of failure due to tensile
rupture
In the analyses to follow, the probability of failure for the
limit state function expressed by Eq. (13) is calculated for the
case of variability in the prediction of installation damage and
variability in the prediction of creep-reduced strength. As
noted earlier, biases in the prediction of strength reductions
related to geogrid durability and junctions in the allowable
load design equation (Eq. (1)) are not considered. Indeed these
statistics are unlikely to be available even if the potential for
these strength loss mechanisms exists. In practice the magni-
tudes of both durability and junction strength reduction factors
should be project specific, but are most often taken as
RFD ¼ RFJ ¼ 1. In the calculations to follow μX D ¼ μX J ¼ 1 and
COVX D ¼ COVX J ¼ 0.
In current design practice the losses of strength in reinforce-
ment geosynthetics due to installation damage and creep are
treated as independent mechanisms. This allows the installa-
tion and creep-reduced strength factors RFID and RFCR to be
multiplied together as shown in Eq. (3). Allen and Bathurst
(1996) concluded that the multiplication of creep and installa-
tion damage reduction factors together (i.e. the conventional
ASD approach) is generally conservative for PET and HDPE
materials and hence results in error on the safe side for design.
Greenwood (2002) carried out stepped isothermal method
(SIM) tests on virgin and damaged geotextile specimens with
exposed polyester fibers. The specimens were damaged by
mechanical abrasion in the laboratory. The “installation”
damage factor was RFID ¼ 1.42 and the creep reduction factor
for undamaged specimens was RFCR ¼ 1.70. Hence the con-
ventional multiplication of these two values yields RFID 
RFCR ¼ 2.41. This value is judged to be reasonably close but
greater than the (combined) creep reduction factor from
damaged specimens of 2.26.
Kongkitkul et al. (2007b) used a non-linear three-component
model for geosynthetic reinforcement to simulate concurrent
degradation and creep mechanisms. Their model showed that
the product RFCR  RFD under-estimates the available tensile
strength of geosynthetic reinforcement after combined creep
and installation damage.
Jeon and Bouazza (2010) investigated the combined effect
of installation damage and creep on 12 different geogrids
(woven PET and uniaxial HDPE). The field damage tests were
carried out in Korea. For the woven geogrids the separate
multiplication of factors resulted in an overestimation of the
combined effect by about 4
20%, and up to about 2% for the
uniaxial HDPE geogrids.
Based on the above prior work the treatment of damage and
creep as independent mechanisms is a practical assumption and
any error is likely small and conservative (i.e. safe) for design.
Assuming that damage and creep bias distributions (XR and
XQ) are lognormal distributed and uncorrelated, the variability
443
in combined bias value X R ¼ X ID X CR (Eq. (9)) is
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
COVR ¼ COV2X ID þ COV2X CR (Eq. (11)). Combined values
are given in Table 3 using maximum bias values for the two
contributing quantities (random variables). In most cases the
installation damage bias term dominates. Where this is not the
case, the combined COV remains very small.
Greenwood (2002) also reported the coefficient of varia-
bility of tensile strength (bias) values for undamaged and
damaged specimens as COVX ult ¼ 0.03 and COVX ID ¼ 0.07,
respectively. The spread in creep bias values for initially
undamaged specimens is COVX CR ¼ 0:02 using the data in his
paper corresponding to times to rupture between 1000 and
106 h (i.e. the range recommended by Bathurst et al. (2012a)).
Greenwood (2002) reported COVR ¼ 0.04 for the creep-
reduced strength of initially damaged specimens which is less
than the value of COVR ¼ 0.07 computed using Eq. (11).
Hence, based on this limited testing it may be expected that the
combined variability due to creep and installation damage may
be over-estimated using Eq. (11) but this error is on the safe
side for design. More research is required to better understand
the synergistic effects of damage and creep on available tensile
strength at the end of design life.
The probability of failure [Pf(g o 0)] as a function of the
ratio F ¼ γQ/φ is plotted in Fig. 2 for the case of two sets of
load bias statistics and a range of COV values for the
combined bias value XR. Fig. 2a is the case of an idealized
good load model for which μQ ¼ 1 and COVQ ¼ 0.30. The
value of COVQ ¼ 0.30 is selected because the load bias
variability cannot be less than the combined variabilities in
the input parameters which are used in the calculation of Tmax
(e.g. friction angle of the soil and unit weight in the calculation
of maximum reinforcement load in a reinforced soil wall
application). The second case is for a load model which is less
accurate with respect to prediction accuracy and has greater
variability in input parameters (average measured to predicted
load μQ ¼ 0.30 and COVQ ¼ 0.50). These bias statistics are
close to those reported by Miyata and Bathurst (2007). They
computed bias values using measured reinforcement loads
from reinforced soil walls in Japan and matching predicted
loads using the PWRC (2000a) method in use at that time.
Similar values have been reported by Bathurst et al. (2012b)
using the AASHTO (2012) Simplified Method. Superimposed
on the two plots are values of F ¼ γQ/φ representing the
minimum values of factor of safety of 1.0 and 1.5 permitted
in Japan (PWRC, 2013) and in the USA (AASHTO, 2012) for
the reinforcement rupture limit state in reinforced soil wall
structures. For a reinforcement layer designed with F ¼ 1 and a
good model, the probability of tensile failure is very large
(  50%) (Fig. 2a). The same F ¼ 1 but using the poorer
model gives lower probabilities of failure because the under-
estimation of the actual loads provides an additional margin of
safety (Fig. 2b).
Both plots show that as the factor of safety goes up, the
probability of failure for any set of resistance bias statistics
goes down. However, the influence of magnitude of spread in
installation damage estimates (bias values) is less critical for
444 R.J. Bathurst, Y. Miyata / Soils and Foundations 55 (2015) 437–446
Fig. 2. Influence of combined installation damage and creep bias variability.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
(COVR ¼ COV2X ID þ COV2X CR ) on probability of failure (Pf) of the reinforce-
ment tensile rupture limit state equation. (a) Reinforcement loads calculated
using good load model (μQ ¼ 1 and COVQ ¼ 0.30). (b) Reinforcement loads
calculated using poor load model (μQ ¼ 0.3 and COVQ ¼0.50).
the poor (i.e. conservative) load model. To achieve Pf ¼ 0.001
(0.1% or 1/1000) using the good load model and combined
COV of resistance bias values COVR ¼ 0.10 requires that the
minimum factor of safety for design must be F ¼ 2.50. For the
poor model case, and the same combined resistance bias COV,
a lower factor safety F¼ 1.2 is satisfactory. The practical
implication of these analyses is that installation damage (regard-
less of additional uncertainty in predicted creep-reduced
strength) is less likely to be a practical issue for poor load
models because these models include a margin of safety due to
conservative load estimates. However, as load model accuracy
improves, the magnitude of installation damage factor becomes
more important to achieve the same margin of safety.
The horizontal axis in the two plots of Fig. 2 is expressed as
F ¼ γQ/φ and thus provides a link between ASD and LRFD
approaches for the tensile rupture limit state. In US practice
(AASHTO, 2012) the load factor for calculation of maximum
load in a reinforcement layer in a geogrid reinforced soil wall
due to soil self-weight is γQ ¼ 1.35 and the resistance factor
is taken as φ ¼ 0.90; this yields F ¼ 1.5. In fact, the choice of
load and resistance factors in LRFD calibration is influenced
by fitting to past practice as demonstrated here. Ideally, for
rigorous LRFD calibration of simple linear state functions of
the type used here, the load factor is selected first based on an
acceptable probability of load exceedance. The resistance
factor is then selected so that Pf(g o0) (Eq. (13)) does not
exceed an acceptable value determined by the developers of
the LRFD code.
The effect of using Tult o T ult;meas at design time (e.g.
manufacturer recommended reference strength value that is
lower than the true mean value) is equivalent to using a mean
of resistance bias μR 41 in Eq. (15). The result is a decrease in
the probability of failure for the same design factor of safety
F ¼ γQ/φ. This can be appreciated from the dashed lines
superimposed in Fig. 2a and b corresponding to μR ¼ 1.35
which is the upper limit of the ratio of T ult;meas /Tult as noted
earlier using the manufacturer recommended strength values in
the study database (Table 3). Using lower reference strength
values reduces the probability of failure associated with the
same COVR value.
6. Conclusions and discussion
The current study has focused on the influence of variability
in the prediction of reduced tensile strength in geogrid
reinforcement products due to installation damage and creep
on the probability of tensile rupture. Variability has been
quantified by bias values computed as the ratio of measured
strength to predicted (nominal) strength values. Bias statistics
have been computed from data reported in PWRC product
certification reports issued in Japan.
The data show that variability in the prediction of tensile
strengths after installation damage is low (less than 14%) for the
26 geogrid products investigated. The variability in prediction of
creep-reduced strength is even lower (7% or less and most often
less than 5%). As-manufactured strength bias values for the
same products are less than 7% and most often less than 3%. It
can be concluded that variability in creep-reduced strength
predictions is largely due to the inherent variability in the
undamaged strength of the material at time of manufacture.
The maximum combined variability on the resistance side of
the limit state function for tensile rupture is less than 14%
using all data and ignoring other strength reduction factors
(durability and junctions). However, the estimate of variability
in bias values for the case of combined installation damage and
creep is likely conservative (i.e. on the safe side for design), at
least for PET reinforcement geosynthetics, based on limited
available testing by Greenwood (2002). The bias statistics
show that in most cases the combined variability due to
installation and creep is dominated by installation damage.
When this is not the case the combined variability is less than
5% and from a practical point of view the resistance side in the
tensile rupture limit state could be treated as deterministic (i.e.
μR ¼ 1, COVR ¼ 0).
On the load side it is reasonable to assume that for an
accurate load model the mean load bias is one and the
coefficient of variability (COVQ) in the prediction of the
maximum tensile load cannot be less than 30%. For a poor
load model the mean of load bias values can be as low as
 R.J. Bathurst, Y. Miyata / Soils and Foundations 55 (2015) 437–446
μQ ¼ 0.30 and the COVQ as great as 50%. The practical
consequence is that probability of failure for tensile rupture
of geogrid products is influenced more by the variability in the
accuracy of load predictions than variability on the resistance
side due to combined variability in the prediction of reduced
tensile strength due to installation damage and creep. As a
result, more accurate load models will require a higher factor
of safety than less accurate load models if both are to achieve
the same probability of failure against tensile rupture. The
reason for this outcome is that the conservative under-
prediction of reinforcement load that occurs with a poor load
model effectively provides an additional margin of safety.
There are however other disadvantages to poor load models.
Clearly they can result in more reinforcement. Also, the
accuracy of a poor load model will typically vary with
magnitude of load prediction (Bathurst et al., 2011a). This load
dependence violates the assumption of uncorrelated load bias
and predicted load that is required to make the transformation
from Eq. (4) to (7). A solution to this problem is to use different
load factors for different reinforcement load ranges (Bathurst
et al., 2008). However, this reduces the utility of any limit state
design equation expressed within a LRFD framework.
Regardless of the accuracy of the load model in the ultimate
(tensile) rupture limit state design equation, the use of
conservative estimates of the reference ultimate strength of a
geogrid reinforcement product (Tult in Eq. (3)) will reduce the
probability of failure for the same design factor of safety.
There are also other unquantifiable margins of safety that
influence the actual tensile load Tmax in any geogrid reinforced
soil application; for example, the tendency by engineers to use
conservative estimates of soil strength parameters for the
reinforced soil. The maximum load in the reinforcement can
also be influenced by quality of construction. In this paper we
have assumed that the quality of construction is uniformly
good and project conditions remain constant over the life of
the structure. Finally the quantitative outcomes reported here
apply only to projects where the installation damage and creep
reduction factors, and spread in bias values, fall within the
envelope of values reported in this paper.
Acknowledgments
The authors are grateful to the Public Works Research
Center, Japan, for permission to use their data for preparation
of this paper, including additional data not available in the
product certification reports. The second author is grateful for
funding awarded by the Japan Ministry of Education, Culture,
Sports, Science and Technology (Grant-in-Aid for Scientific
Research (B) no. 24360195) and the Japan Ministry of Defense
to collaborate with the first author in Canada where the
preparation of this paper was carried out.
References
AASHTO, 2012. AASHTO LRFD Bridge Design Specifications, sixth edition
American Association of State Highway and Transportation Officials,
Washington, DC, USA.
445
ASTM D6637, 2001. Standard Test Method for Determining Tensile Proper-
ties of Geogrids by the Single or Multi-Rib Tensile Method. American
Society for Testing and Materials, West Conshohocken, PA, USA.
Akagi, T., Chida, S., Yamamoto, C., Miki, H., 2004. PWRC certification of
geosynthetics. In: Proceedings of the 3rd Asian Regional Conference on
Geosynthetics, pp. 363
368.
Allen, T.M., Bathurst, R.J., 1996. Investigation of the combined allowable
strength reduction factor for geosynthetic creep and installation damage.
Geosynth. Int. 3 (3), 407–439.
Allen, T.M., Nowak, A.S., Bathurst, R.J., 2005. Calibration to Determine Load
and Resistance Factors for Geotechnical and Structural Design. Transpor-
tation Research Board Circular E-C079, Washington, DC, 93 pp.
Ang, A.H.-S., Tang, W.H., 1975. Probability Concepts in Engineering
Planning and Design, Vol. 1
Basic Principles; 2
Decision, Risk,
and Reliability. John Wiley & Sons, New York, NY, USA.
Bathurst, R.J., 2014. LRFD calibration of simple limit state functions in
geotechnical soil-structure design. In: Phoon, K.K., Ching, J. (Eds.),
Reliability-Based Design in Geotechnical Engineering: Computations and
Applications. Spon Press, Oxon, UK, pp. 339–354 (Chapter 8).
Bathurst, R.J., Allen, T.M., Nowak, A.S., 2008. Calibration concepts for load
and resistance factor design (LRFD) of reinforced soil walls. Can. Geotech.
J. 45 (10), 1377–1392.
Bathurst, R.J., Cai, Z., 1994. In-isolation cyclic load
extension behavior of
two geogrids. Geosynth. Int. 1 (1), 3–17.
Bathurst, R.J., Huang, B., Allen, T.M., 2011a. Load and resistance factor
design (LRFD) calibration for steel grid reinforced soil walls. Georisk 5
(3
4), 218–228.
Bathurst, R.J., Huang, B., Allen, T.M., 2012a. Interpretation of laboratory
creep testing for reliability-based analysis and load and resistance factor
design (LRFD) calibration. Geosynth. Int. 19 (1), 39–53.
Bathurst, R.J., Huang, B., Allen, T.M., 2012b. LRFD calibration of the
ultimate pullout limit state for geogrid reinforced soil retaining walls.
ASCE Int. J. Geomech. 12 (4), 399–413 (Special Issue on Geosynthetics).
Bathurst, R.J., Miyata, Y., Konami, T., 2011b. Limit states design calib-
ration for internal stability of multi-anchor walls. Soils Found. 51 (6), 1051–1064.
Bathurst, R.J., Huang, B., Allen, T.M., 2011c. Analysis of installation damage
tests for LRFD calibration of reinforced soil structures. Geotext. Geo-
membr. 29 (3), 323–334.
CSA, 2006. Canadian Highway Bridge Design Code (CHBDC). CSA Standard
S6-06. Canadian Standards Association (CSA), Toronto, Ontario.
Greenwood, J.H., 2002. The effect of installation damage on the long term design
strength of a reinforcing geosynthetic. Geosynth. Int. 9 (3), 247–258.
Hirakawa, D., Kongkitkul, W., Tatsuoka, F., Uchimura, T., 2003. Time-
dependent stress
strain behaviour due to viscous properties of geogrid
reinforcement. Geosynth. Int. 10 (6), 176–199.
Huang, B., Bathurst, R.J., 2009. Evaluation of soil-geogrid pullout models
using a statistical approach. ASTM Geotech. Test. J. 32 (6), 489–504.
Kongkitkul, W., Hirakawa, D., Tatsuoka, F., Uchimura, T., 2004. Viscous
deformation of geosynthetic reinforcement under cyclic loading conditions
and its model simulation. Geosynth. Int. 11 (2), 73–99.
Kongkitkul, W., Hirakawa, D., Tatsuoka, F., 2007a. Viscous behavior of
geogrids; experiment and simulation. Soils Found. 47 (2), 265–283.
Kongkitkul, W., Tatsuoka, F., Hirakawa, D., 2007b. Creep rupture curve for
simultaneous creep deformation and degradation of geosynthetic reinforce-
ment. Geosynth. Int. 14 (4), 189–200.
Jeon, H.Y., Bouazza, A., 2010. Experimental investigation of installation
damage for geogrids. Ground Improv. 163 (4), 197–205.
Miyata, Y., Bathurst, R.J., 2007. Evaluation of K-Stiffness method for vertical
geosynthetic reinforced granular soil walls in Japan. Soils Found. 47 (2), 319–335.
Miyata, Y., Bathurst, R.J., 2012. Reliability analysis of soil-geogrid pullout
models in Japan. Soils Found. 52 (4), 620–633.
Miyata, Y., Bathurst, R.J., 2015. Reliability analysis of installation damage test
data in Japan. Soils Found. 55 (2), 393–403.
Miyata, Y., Bathurst, R.J., Allen, T.M., 2014. Reliability analysis of geogrid
creep data in Japan. Soils Found. 54 (4), 608–620.
PWRC, 2000a. Design and Construction Manual of Geosynthetics Reinforced
Soil (revised version). Public Works Research Center, Tsukuba, Ibaraki,
Japan, 305 pp. (in Japanese).
446 R.J. Bathurst, Y. Miyata / Soils and Foundations 55 (2015) 437–446
PWRC, 2000b. Technical Report for PWRC Certification of Construction
Material and Technique: Spike Grid Certificate No. 1110. Public Works
Research Center, Tsukuba, Ibaraki, Japan, 135 pp. (in Japanese).
PWRC, 2002b. Technical Report for PWRC Certification of Construction
Material, Products and Technique: Adem Type-G Certificate No. 0120. 2nd
edition Public Works Research Center, Tsukuba, Ibaraki, Japan,
171 pp. (in Japanese).
PWRC, 2004a. Technical Report for PWRC Certification of Construction
Material, Products and Technique: Power Grid Certificate No. 0412. Public
Works Research Center, Tsukuba, Ibaraki, Japan, 169 pp. (in Japanese).
PWRC, 2004b. Technical Report for PWRC Certification of Construction
Material, Products and Technique: Tenax Certificate No. 0413. Public
Works Research Center, Tsukuba, Ibaraki, Japan, 150 pp. (in Japanese).
PWRC, 2006. Technical Report for PWRC Certification of Construction
Material, Products and Technique: Geogrid Tensar Certificate No. 0901.
2nd edition Public Works Research Center, Tsukuba, Ibaraki, Japan,
156 pp. (in Japanese).
PWRC, 2007a. Technical Report for PWRC Certification of Construction
Material, Products and Technique: Cell Force Type-F Certificate No. 0214.
2nd edition Public Works Research Center, Tsukuba, Ibaraki, Japan,
118 pp. (in Japanese).
PWRC, 2008a. Technical Report for PWRC Certification of Construction
Material, Products and Technique: Adem Type-HG Certificate No. 0804.
Public Works Research Center, Tsukuba, Ibaraki, Japan,
115 pp. (in Japanese).
PWRC, 2008b. Technical Report for PWRC Certification of Construction
Material, Products and Technique: Trigrid Certificate No. 0808. Public
Works Research Center, Tsukuba, Ibaraki, Japan, 113 pp. (in Japanese).
PWRC, 2013. Design and Construction Manual of Geosynthetics Reinforced
Soil (second revision). Public Works Research Center, Tsukuba, Ibaraki,
Japan, 305 pp. (in Japanese).
Shinoda, M., Bathurst, R.J., 2004a. Lateral and axial deformation of PP, HDPE
and PET geogrids under tensile load. Geotext. Geomembr. 22 (4),
205–222.
Shinoda, M., Bathurst, R.J., 2004b. Strain measurement of geogrids using a
video-extensometer technique. ASTM Geotech. Test. J. 27 (5), 456–463.
Shinoda, M., Bathurst, R.J., Tatsuoka, F., Horii, K., 2002. Time-dependent
deformation and strength characteristics of geogrids due to viscous
properties. Japan Chapter of IGS, vol. 17; pp. 137–144.
Tanaka, T., Sakai, T., 1979. A theoretical study on the fatigue life distribution
of metallic materials near the fatigue limit. Trans. Jpn. Soc. Mech. Eng.
A45 (391), 211–219.
Walters, D.L., Allen, T.M., Bathurst, R.J., 2002. Conversion of geosynthetic
strain to load using reinforcement stiffness. Geosynth. Int. 9 (5
6),
483–523.